基于LSTM水稻生育期地下水水位预测——以黑龙江省查哈阳灌区为例  

Prediction of groundwater level based on LSTM——Taking Chahayang Irrigation District in Heilongjiang Province as an example

作　　者：徐淑琴 乔厚清 王雅君 李仲裕 郭晓婷 XU Shuqin;QIAO Houqing;WANG Yajun;LI Zhongyu;GUO Xiaoting(School of Water Conservancy and Civil Engineering,Northeast Agricultural University,Harbin 150030,China)

机构地区：[1]东北农业大学水利与土木工程学院,哈尔滨150030

出　　处：《东北农业大学学报》2021年第2期70-78,共9页Journal of Northeast Agricultural University

基　　金：国家重点研发计划项目(2017YFC0404503)。

摘　　要：为解决以往基于链式神经网络水文时间序列预测模型无法较好实现长依赖预测局限性问题,探讨长短期记忆神经网络结构,重点分析"门"运作模式与其解决长依赖问题原理,以黑龙江省查哈阳灌区为例,采用单变量时间序列建立基长短期神经网络地下水水位预测模型,预测该区2019~2036年地下水水位。对比预测结果与实测值均值、标准差,并用快速傅里叶变换计算幅度谱,取得200与400 Hz较优频率。结果表明,模型预测结果与实测值周期性、变化趋势及数值范围接近,模型预测效果较优,可用作查哈阳灌区水稻生育期地下水水位预测。In order to solve the problem that the previous hydrological time series forecasting model based on chain neural network cannot well realize the long-term dependence problem prediction,this study discusses the structure of long short term memory networks,focusing on its"gate"operation mode and its solution to long-term dependence problem.The principle was analyzed,and the Chahayang Irrigation District in Heilongjiang Province was taken as an example,a univariate time series was used to establish a groundwater level pre-model based on a long short term memory networks to predict the groundwater level in the district from 2019 to 2036.The prediction result was compared with the mean value and standard deviation of the measured value,and the amplitude spectrum was obtained by fast Fourier transform,and the better frequency of 200 and 400 Hz was obtained.The results showed that the model prediction results were similar to the measured values in terms of periodicity,changing trend and numerical range,and the model prediction effect was relatively good.It could be used to predict the groundwater level during the rice growth period in Chahayang Irrigation District.

关 键 词：灌溉水资源 地下水水位 长短期记忆神经网络 幅度谱 

分 类 号：S277[农业科学—农业水土工程]